Graph-of-Mark: Promote Spatial Reasoning in Multimodal Language Models with Graph-Based Visual Prompting
This addresses the limitation of existing visual prompting methods that treat objects as isolated, improving spatial reasoning for multimodal language models in tasks like visual question answering, though it is incremental as it builds on prior techniques like Set-of-Mark.
The paper tackled the problem of enhancing spatial reasoning in multimodal language models by proposing Graph-of-Mark, a graph-based visual prompting technique that overlays scene graphs onto images to capture object relationships, resulting in improvements of up to 11 percentage points in accuracy for visual question answering and localization tasks.
Recent advances in training-free visual prompting, such as Set-of-Mark, have emerged as a promising direction for enhancing the grounding capabilities of multimodal language models (MLMs). These techniques operate by partitioning the input image into object regions and annotating them with marks, predominantly boxes with numeric identifiers, before feeding the augmented image to the MLM. However, these approaches treat marked objects as isolated entities, failing to capture the relationships between them. On these premises, we propose Graph-of-Mark (GoM), the first pixel-level visual prompting technique that overlays scene graphs onto the input image for spatial reasoning tasks. We evaluate GoM across 3 open-source MLMs and 4 different datasets, conducting extensive ablations on drawn components and investigating the impact of auxiliary graph descriptions in the text prompt. Our results demonstrate that GoM consistently improves the zero-shot capability of MLMs in interpreting object positions and relative directions, improving base accuracy in visual question answering and localization up to 11 percentage points.